Event-related potentials (ERPs) were recorded as subjects read semantically meaningful, syntactically legal but nonsensical and random word strings. The constraints imposed by formal sentence structure alone did not reduce the amplitude of the N400 component elicited by open-class words, whereas semantic constraints did. Semantic constraints also eliminated the word-frequency effect of a larger N400 for low-frequency words. Responses to closed-class words exhibited reduced N400 amplitudes in syntactic and congruent sentences, indicating that formal sentence structure placed greater restrictions on closed-class words than it did on open-class words. However, unlike the open-class results, the impact of sentence context on closed-class words was stable across word positions, suggesting that these syntactic constraints were applied only locally. A second ERP component, distinct from the N400, was elicited primarily by congruent closed-class words.Much experimental effort has been devoted to demonstrating that people are faster at identifying (or require less sensory information to identify) words that occur in an informative context than they are at identifying words that do not. In natural discourse, readers and listeners have several theoretically distinct sources of information that they may use in the transformation of a visual or acoustic signal into a meaningful representation. These include associations between individual words in long-term memory, semantic information derived from larger chunks of the ongoing discourse or real-world knowledge, syntactic restrictions provided by the grammar of the language, the overall probability of occurrence of particular words based on their frequency of usage in the reader's/listener's experience with the language, and pragmatic cues. Psycholinguistic research has attempted to describe how each source of information is used, addressing such questions as whether different cues are utilized serially or in parallel, whether they are allowed to interact or are kept distinct until a relatively late stage of processing, whether each type of information is applied to all words or only a subset, and whether or not different sources of information or vocabulary types are interpreted by distinct neural circuits.In the present experiment, we examined the impact of three potentially informative cues for the processing of A brief report of this study was presented at the 28th annual meeting of the Society for Psychophysiological Research, in October 1988. We are grateful to Jon Hansen for software support, to Jeff Elman. Liz Bates. Robert Proctor, and two anonymous reviewers for commenting on an earlier version of the paper, to Cathy Harris for helpful discussions, and to Steve Hillyard for making his facilities available for preparation of the manuscript.